xref: /freebsd/cddl/usr.sbin/zfsd/zfsd.cc (revision f374ba41)
1 /*-
2  * Copyright (c) 2011, 2012, 2013, 2014, 2015, 2016  Spectra Logic Corporation
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions, and the following disclaimer,
10  *    without modification.
11  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
12  *    substantially similar to the "NO WARRANTY" disclaimer below
13  *    ("Disclaimer") and any redistribution must be conditioned upon
14  *    including a substantially similar Disclaimer requirement for further
15  *    binary redistribution.
16  *
17  * NO WARRANTY
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
21  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGES.
29  *
30  * Authors: Justin T. Gibbs     (Spectra Logic Corporation)
31  */
32 
33 /**
34  * \file zfsd.cc
35  *
36  * The ZFS daemon consumes kernel devdctl(4) event data via devd(8)'s
37  * unix domain socket in order to react to system changes that impact
38  * the function of ZFS storage pools.  The goal of this daemon is to
39  * provide similar functionality to the Solaris ZFS Diagnostic Engine
40  * (zfs-diagnosis), the Solaris ZFS fault handler (zfs-retire), and
41  * the Solaris ZFS vdev insertion agent (zfs-mod sysevent handler).
42  */
43 
44 #include <sys/cdefs.h>
45 #include <sys/byteorder.h>
46 #include <sys/param.h>
47 #include <sys/fs/zfs.h>
48 
49 #include <err.h>
50 #include <fcntl.h>
51 #include <libgeom.h>
52 #include <libutil.h>
53 #include <poll.h>
54 #include <syslog.h>
55 
56 #include <libzfs.h>
57 
58 #include <list>
59 #include <map>
60 #include <string>
61 
62 #include <devdctl/guid.h>
63 #include <devdctl/event.h>
64 #include <devdctl/event_factory.h>
65 #include <devdctl/exception.h>
66 #include <devdctl/consumer.h>
67 
68 #include "callout.h"
69 #include "vdev_iterator.h"
70 #include "zfsd_event.h"
71 #include "case_file.h"
72 #include "vdev.h"
73 #include "vdev_iterator.h"
74 #include "zfsd.h"
75 #include "zfsd_exception.h"
76 #include "zpool_list.h"
77 
78 __FBSDID("$FreeBSD$");
79 
80 /*================================== Macros ==================================*/
81 #define NUM_ELEMENTS(x) (sizeof(x) / sizeof(*x))
82 
83 /*============================ Namespace Control =============================*/
84 using DevdCtl::Event;
85 using DevdCtl::EventFactory;
86 using DevdCtl::EventList;
87 
88 /*================================ Global Data ===============================*/
89 int              g_debug = 0;
90 libzfs_handle_t *g_zfsHandle;
91 
92 /*--------------------------------- ZfsDaemon --------------------------------*/
93 //- ZfsDaemon Static Private Data ----------------------------------------------
94 ZfsDaemon	    *ZfsDaemon::s_theZfsDaemon;
95 bool		     ZfsDaemon::s_logCaseFiles;
96 bool		     ZfsDaemon::s_terminateEventLoop;
97 char		     ZfsDaemon::s_pidFilePath[] = "/var/run/zfsd.pid";
98 pidfh		    *ZfsDaemon::s_pidFH;
99 int		     ZfsDaemon::s_signalPipeFD[2];
100 bool		     ZfsDaemon::s_systemRescanRequested(false);
101 EventFactory::Record ZfsDaemon::s_registryEntries[] =
102 {
103 	{ Event::NOTIFY, "GEOM",  &GeomEvent::Builder },
104 	{ Event::NOTIFY, "ZFS",   &ZfsEvent::Builder }
105 };
106 
107 //- ZfsDaemon Static Public Methods --------------------------------------------
108 ZfsDaemon &
109 ZfsDaemon::Get()
110 {
111 	return (*s_theZfsDaemon);
112 }
113 
114 void
115 ZfsDaemon::WakeEventLoop()
116 {
117 	write(s_signalPipeFD[1], "+", 1);
118 }
119 
120 void
121 ZfsDaemon::RequestSystemRescan()
122 {
123 	s_systemRescanRequested = true;
124 	ZfsDaemon::WakeEventLoop();
125 }
126 
127 void
128 ZfsDaemon::Run()
129 {
130 	ZfsDaemon daemon;
131 
132 	while (s_terminateEventLoop == false) {
133 
134 		try {
135 			daemon.DisconnectFromDevd();
136 
137 			if (daemon.ConnectToDevd() == false) {
138 				sleep(30);
139 				continue;
140 			}
141 
142 			daemon.DetectMissedEvents();
143 
144 			daemon.EventLoop();
145 
146 		} catch (const DevdCtl::Exception &exp) {
147 			exp.Log();
148 		}
149 	}
150 
151 	daemon.DisconnectFromDevd();
152 }
153 
154 //- ZfsDaemon Private Methods --------------------------------------------------
155 ZfsDaemon::ZfsDaemon()
156  : Consumer(/*defBuilder*/NULL, s_registryEntries,
157 	    NUM_ELEMENTS(s_registryEntries))
158 {
159 	if (s_theZfsDaemon != NULL)
160 		errx(1, "Multiple ZfsDaemon instances created. Exiting");
161 
162 	s_theZfsDaemon = this;
163 
164 	if (pipe(s_signalPipeFD) != 0)
165 		errx(1, "Unable to allocate signal pipe. Exiting");
166 
167 	if (fcntl(s_signalPipeFD[0], F_SETFL, O_NONBLOCK) == -1)
168 		errx(1, "Unable to set pipe as non-blocking. Exiting");
169 
170 	if (fcntl(s_signalPipeFD[1], F_SETFL, O_NONBLOCK) == -1)
171 		errx(1, "Unable to set pipe as non-blocking. Exiting");
172 
173 	signal(SIGHUP,  ZfsDaemon::RescanSignalHandler);
174 	signal(SIGINFO, ZfsDaemon::InfoSignalHandler);
175 	signal(SIGINT,  ZfsDaemon::QuitSignalHandler);
176 	signal(SIGTERM, ZfsDaemon::QuitSignalHandler);
177 	signal(SIGUSR1, ZfsDaemon::RescanSignalHandler);
178 
179 	g_zfsHandle = libzfs_init();
180 	if (g_zfsHandle == NULL)
181 		errx(1, "Unable to initialize ZFS library. Exiting");
182 
183 	Callout::Init();
184 	InitializeSyslog();
185 	OpenPIDFile();
186 
187 	if (g_debug == 0)
188 		daemon(0, 0);
189 
190 	UpdatePIDFile();
191 }
192 
193 ZfsDaemon::~ZfsDaemon()
194 {
195 	PurgeCaseFiles();
196 	ClosePIDFile();
197 }
198 
199 void
200 ZfsDaemon::PurgeCaseFiles()
201 {
202 	CaseFile::PurgeAll();
203 }
204 
205 bool
206 ZfsDaemon::VdevAddCaseFile(Vdev &vdev, void *cbArg)
207 {
208 	if (vdev.State() != VDEV_STATE_HEALTHY)
209 		CaseFile::Create(vdev);
210 
211 	return (/*break early*/false);
212 }
213 
214 void
215 ZfsDaemon::BuildCaseFiles()
216 {
217 	ZpoolList zpl;
218 	ZpoolList::iterator pool;
219 
220 	/* Add CaseFiles for vdevs with issues. */
221 	for (pool = zpl.begin(); pool != zpl.end(); pool++)
222 		VdevIterator(*pool).Each(VdevAddCaseFile, NULL);
223 
224 	/* De-serialize any saved cases. */
225 	CaseFile::DeSerialize();
226 
227 	/* Simulate config_sync events to force CaseFile reevaluation */
228 	for (pool = zpl.begin(); pool != zpl.end(); pool++) {
229 		char evString[160];
230 		Event *event;
231 		nvlist_t *config;
232 		uint64_t poolGUID;
233 		const char *poolname;
234 
235 		poolname = zpool_get_name(*pool);
236 		config = zpool_get_config(*pool, NULL);
237 		if (config == NULL) {
238 			syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
239 			    "find pool config for pool %s", poolname);
240 			continue;
241 		}
242 		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
243 				     &poolGUID) != 0) {
244 			syslog(LOG_ERR, "ZFSDaemon::BuildCaseFiles: Could not "
245 			    "find pool guid for pool %s", poolname);
246 			continue;
247 		}
248 
249 
250 		snprintf(evString, 160, "!system=ZFS subsystem=ZFS "
251 		    "type=misc.fs.zfs.config_sync sub_type=synthesized "
252 		    "pool_name=%s pool_guid=%" PRIu64 "\n", poolname, poolGUID);
253 		event = Event::CreateEvent(GetFactory(), string(evString));
254 		if (event != NULL) {
255 			event->Process();
256 			delete event;
257 		}
258 	}
259 }
260 
261 void
262 ZfsDaemon::RescanSystem()
263 {
264         struct gmesh	  mesh;
265         struct gclass	 *mp;
266         struct ggeom	 *gp;
267         struct gprovider *pp;
268 	int		  result;
269 
270         /*
271 	 * The devdctl system doesn't replay events for new consumers
272 	 * of the interface.  Emit manufactured DEVFS arrival events
273 	 * for any devices that already before we started or during
274 	 * periods where we've lost our connection to devd.
275          */
276 	result = geom_gettree(&mesh);
277 	if (result != 0) {
278 		syslog(LOG_ERR, "ZfsDaemon::RescanSystem: "
279 		       "geom_gettree failed with error %d\n", result);
280 		return;
281 	}
282 
283 	const string evStart("!system=DEVFS subsystem=CDEV type=CREATE "
284 			     "sub_type=synthesized cdev=");
285         LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
286                 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
287                         LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
288 				Event *event;
289 
290 				string evString(evStart + pp->lg_name + "\n");
291 				event = Event::CreateEvent(GetFactory(),
292 							   evString);
293 				if (event != NULL) {
294 					if (event->Process())
295 						SaveEvent(*event);
296 					delete event;
297 				}
298                         }
299                 }
300 	}
301 	geom_deletetree(&mesh);
302 }
303 
304 void
305 ZfsDaemon::DetectMissedEvents()
306 {
307 	do {
308 		PurgeCaseFiles();
309 
310 		/*
311 		 * Discard any events waiting for us.  We don't know
312 		 * if they still apply to the current state of the
313 		 * system.
314 		 */
315 		FlushEvents();
316 
317 		BuildCaseFiles();
318 
319 		/*
320 		 * If the system state has changed during our
321 		 * interrogation, start over.
322 		 */
323 	} while (s_terminateEventLoop == false && EventsPending());
324 
325 	RescanSystem();
326 }
327 
328 void
329 ZfsDaemon::EventLoop()
330 {
331 	while (s_terminateEventLoop == false) {
332 		struct pollfd fds[2];
333 		int	      result;
334 
335 		if (s_logCaseFiles == true) {
336 			EventList::iterator event(m_unconsumedEvents.begin());
337 			s_logCaseFiles = false;
338 			CaseFile::LogAll();
339 			while (event != m_unconsumedEvents.end())
340 				(*event++)->Log(LOG_INFO);
341 		}
342 
343 		Callout::ExpireCallouts();
344 
345 		/* Wait for data. */
346 		fds[0].fd      = m_devdSockFD;
347 		fds[0].events  = POLLIN;
348 		fds[0].revents = 0;
349 		fds[1].fd      = s_signalPipeFD[0];
350 		fds[1].events  = POLLIN;
351 		fds[1].revents = 0;
352 		result = poll(fds, NUM_ELEMENTS(fds), /*timeout*/INFTIM);
353 		if (result == -1) {
354 			if (errno == EINTR)
355 				continue;
356 			else
357 				err(1, "Polling for devd events failed");
358 		} else if (result == 0) {
359 			errx(1, "Unexpected result of 0 from poll. Exiting");
360 		}
361 
362 		if ((fds[0].revents & POLLIN) != 0)
363 			ProcessEvents();
364 
365 		if ((fds[1].revents & POLLIN) != 0) {
366 			static char discardBuf[128];
367 
368 			/*
369 			 * This pipe exists just to close the signal
370 			 * race.  Its contents are of no interest to
371 			 * us, but we must ensure that future signals
372 			 * have space in the pipe to write.
373 			 */
374 			while (read(s_signalPipeFD[0], discardBuf,
375 				    sizeof(discardBuf)) > 0)
376 				;
377 		}
378 
379 		if (s_systemRescanRequested == true) {
380 			s_systemRescanRequested = false;
381 			syslog(LOG_INFO, "System Rescan request processed.");
382 			RescanSystem();
383 		}
384 
385 		if ((fds[0].revents & POLLERR) != 0) {
386 			syslog(LOG_INFO, "POLLERROR detected on devd socket.");
387 			break;
388 		}
389 
390 		if ((fds[0].revents & POLLHUP) != 0) {
391 			syslog(LOG_INFO, "POLLHUP detected on devd socket.");
392 			break;
393 		}
394 	}
395 }
396 //- ZfsDaemon staic Private Methods --------------------------------------------
397 void
398 ZfsDaemon::InfoSignalHandler(int)
399 {
400 	s_logCaseFiles = true;
401 	ZfsDaemon::WakeEventLoop();
402 }
403 
404 void
405 ZfsDaemon::RescanSignalHandler(int)
406 {
407 	RequestSystemRescan();
408 }
409 
410 void
411 ZfsDaemon::QuitSignalHandler(int)
412 {
413 	s_terminateEventLoop = true;
414 	ZfsDaemon::WakeEventLoop();
415 }
416 
417 void
418 ZfsDaemon::OpenPIDFile()
419 {
420 	pid_t otherPID;
421 
422 	s_pidFH = pidfile_open(s_pidFilePath, 0600, &otherPID);
423 	if (s_pidFH == NULL) {
424 		if (errno == EEXIST)
425 			errx(1, "already running as PID %d. Exiting", otherPID);
426 		warn("cannot open PID file");
427 	}
428 }
429 
430 void
431 ZfsDaemon::UpdatePIDFile()
432 {
433 	if (s_pidFH != NULL)
434 		pidfile_write(s_pidFH);
435 }
436 
437 void
438 ZfsDaemon::ClosePIDFile()
439 {
440 	if (s_pidFH != NULL)
441 		pidfile_remove(s_pidFH);
442 }
443 
444 void
445 ZfsDaemon::InitializeSyslog()
446 {
447 	openlog("zfsd", LOG_NDELAY, LOG_DAEMON);
448 }
449 
450